Code-free AI Production Lines

More than 4,500 operational satellites are orbiting the earth, and more than 600 are re-gu-larly taking pictures from the earth’s surface (2018). Numbers are continuously in-creasing.

Synchronously, analysis of high-resolution imagery data with AI is becoming in-creasingly important for business and humanitarian environments:

From detecting buildings and roads to create geographic maps of data-scarce regions of the world for effective emer-gence response, to silo bag detection as re-liable signal of grain production, to many more use cases.

The combination of imagery data and deep & machine learning is on the rise.

Focusing on algorithms is not enough

With the increase of these AI applications, we observe a shift from “one-step” to more complex “multi-step” applications. Think of a yield-related agricultural question:

Computing vegetation indices from drone or satellite images with deep neural networks creates only a line of features and is not enough for yield prediction. More features from other data sources like weather and soil-specific data must be included.

And if we augment these features with prices and even market sentiments, it is possible to compute outstanding and valuable forecasts - with “multi-step” AI applications.

Taking full advantage of imagery data requires to readjust our focus from single or even hybrid algorithms to the architecture of the entire AI triad: data, algorithms, and solution.

AI never becomes business ready with hand-made applications

Todays’ business environments and markets are volatile and ever-changing. Time is the ultimate performance indicator when it comes to data-driven decision making or turning ideas into innovative products and services.

AI can be the key: Provided AI applications can be built on demand and in time, based on a continuous production process. Why? Nowadays data and models are perishable goods, and AI applications that come too late produce costs but no business value.

So today, it is no longer a challenge to define an AI algorithm. The real challenge is about the fast production and deployment of its associated business application - on demand and again and again.

Astraea meets Intel or RasterFrames goes Analytics-Zoo

Moving from hand-made AI projects to completely configurable & code-free AI production lines no longer is a dream:

PredictiveWorks. is proud to announce that imagery data from drones and satellites are fully supported now and can be analyzed with 150+ machine intelligence operators without writing a single line of code.

Closing the imagery data gap is a huge benefit for those who want to process earth observation data with code-free AI recipes and workflows.

This success has never been possible without the outstanding work of others:

Astraea’s release of RasterFrames (the current version is 0.9.1) is one of the building blocks of our success. RasterFrames provides a DataFrame-centric view over arbitrary Earth-Observation data and seamlessly integrates into the Apache Spark ecosystem.

Whatever data is needed to be analyzed; they can be handled like any other data. With RasterFrames integrated into PredictiveWorks., imagery data can now be processed with the deep learning power of Intel's Analytics-Zoo and the machine learning capability of Apache Spark MLlib.

Re-envision AI production with PredictiveWorks.

Multi-step AI applications that cover business and production data, soil and weather data, and aerial and satellite imagery data can now be generated like cars in a production line.

SatelliteWorks. is a brand new AI product that has been built with PredictiveWorks. in less than 4 weeks. For us, this is the new key performance indicator to produce AI applications.